Local stereo matching based on the improved matching cost function and the adaptive window

This paper presents a stereo matching algorithm based on the improved matching cost function and the adaptive window. First of all, AD, Census and gradient gap of left and right images are calculated. Joint bilateral filter is used for calculating weights of pixels, thereby constructing a matching cost function. Then, different aggregation window construction strategies are used for aggregating matching costs through pixel classification. One judgment of disparity computation validness is increased on the basis of using Winner-Takes-All strategy. Finally, disparity map is refined through regional voting, joint image inpainting and other optimization steps. Middlebury data sets are adopted for the experiment. The result shows that algorithm proposed in the paper has good processing result.